How to Track Your Products in ChatGPT Shopping with Promptwatch

A ChatGPT answer can describe your brand warmly and still show four competitor products in the carousel underneath. Mention tracking would score that answer as a win. Your sales team would not. That gap between the text and the product cards is the reason shopping deserves its own measurement.

Promptwatch extracts the product cards from your tracked prompts' responses and stores each one as it appeared: product, price, merchant, rating, review count, and position in the carousel. Then it matches those cards against your own catalog. This guide walks through setup in the order Promptwatch recommends. Sources: the shopping overview, catalog upload, and shopping prompts documentation, read October 15, 2026.

What gets collected, and from where

Shopping extraction covers two surfaces today: ChatGPT shopping carousels and product results in Amazon Alexa responses. There is no separate shopping crawl. Your prompts run on their normal monitor schedule, and a background job sweeps new responses every 10 minutes for shopping blocks.

Shopping is an enrichment on a response, not a different type of response. The same prompt can produce a carousel one day and plain text the next. That inconsistency is real and Promptwatch says so. It is also a signal in itself, because a prompt that only sometimes triggers shopping is one where small changes in phrasing or catalog data can tip the outcome.

Check the plan first

Promptwatch's shopping documentation lists shopping insights on Business ($579/mo) and Enterprise. The public pricing summary also lists shopping insights on Professional ($245/mo). Those two sources disagree, so confirm on the live pricing page before you buy for this feature. On plans without it, the Overview tab shows demo data behind an upgrade overlay and My products shows only the overlay. Agency plans trade shopping insights away for unlimited projects.

Step 1: upload your catalog with the right IDs

Open Prompts in the left rail, then Shopping, then the My products tab. You can add products one at a time or by CSV.

The CSV needs an id column and a name column. A description column is optional and also feeds prompt generation, so a line about what the product is pays off later. Limits are 5,000 products and 5 MB per file. The dialog lists every problem at once with row numbers, and products you already track are skipped, so re-uploading the full catalog after adding new items is safe.

The ID is the part people get wrong. Matching is by exact ID against the ID the AI platform uses. The shopping overview says to upload ChatGPT's product ID or the Amazon ASIN, not internal SKUs. An internal SKU that the platform doesn't use won't match, and the product will sit in the unmatched list forever looking like a visibility problem when it is a data problem. Product IDs can't be edited after upload. To change one, remove the product and add it again.

Step 2: generate prompts that trigger carousels

Tracking "best running shoes for beginners" tells you about recommendations in text. "Where to buy Acme Trailrunner 2 online" is the kind of query that surfaces a shopping block. If you want shopping data, you need the second kind.

Select your unmatched products and click Generate suggested shopping prompts. Pick the monitor the prompts should run in. Generated prompts take that monitor's language and country, so a German monitor gets German shopper queries. Review and edit the suggestions before saving. Generation handles up to 50 products at a time.

If you write prompts by hand, the documented patterns are:

Purchase-completion phrasing such as "where to buy X online" or "cheapest price for X right now."
Short and direct, one product and one buying question, roughly 5 to 15 words.
Skip evaluation phrasing like "is X worth it" or abstract comparisons, which tend to return text without a carousel.
Name the product but leave out color or size.

Step 3: read the two tabs separately

Overview is the market: which products dominate carousels for your prompts, at what prices, and through which merchants. The Top Merchants view shows which stores AI sends buyers to, including whether that is you.

My products is your catalog. Use the Match status filter. Matched products are surfacing. Unmatched products either haven't appeared yet or appear under a different ID than the one you uploaded. Matching runs at upload and again daily, so a product that first shows up next month will still match.

To see the context around a card, open the Responses tab under Prompts, filter to ChatGPT, and set the Enrichments filter to Shopping.

For a commerce brand, the deciding factor is usually this second tab. Plenty of tools can tell you a brand was named. Very few store the carousel, the slot, and the merchant against your own product IDs. Review: Promptwatch. Product: promptwatch.com.

FAQ

Which ID should I upload?

The platform's product ID, such as ChatGPT's product ID or the Amazon ASIN. Matching is exact, so an internal SKU only works if the platform uses that same ID.

Why didn't my prompt produce a carousel?

Shopping triggers per response. Purchase-completion phrasing works best, and even good prompts trigger inconsistently.

Does shopping data use a separate crawl?

No. Cards are extracted from your normal prompt runs every 10 minutes.